China's Tianjic Chip Rides A Bike
Written by Mike James   
Sunday, 11 August 2019

A bike that follows you around? Sounds good and it's not just self-driving it is also self-balancing. Can spiking neuromorphic networks be the future? Researchers in China seem to think so.

The AI powerhouse is generally reckoned to the USA, but China has ambitions and is big enough to realize them. For example, its Tianjic chip is an example of a neuromorphic implementation.

Neuromorphic means that it is like a biological neural network. If you are under the impression that the neural networks we currently use are like biological networks, then think again. The best that can be said is that they are biologically inspired. A neuron gets excited by signals it receives from other neurons and eventually it "fires", passing the signal on. Our current artificial neural networks simply sum the inputs and output an activation signal that is proportional. A neuromorphic network usually copies the way that real neurons behave more closely than this. In this case the artificial neurons receive pulses or spikes from the other networks and when the input is enough of a stimulus it starts to fire spikes to neurons connected to it.

You might ask why we are interested in neuromorphic approaches if our current neural networks are doing so well. The answer is that neuromorphic chips have the promise of being more powerful while using less power. The only real problem is that if you take a direct approach to implementing them - i.e. model neuronal behavior in electronics - you end up with something that is difficult to program and difficult to train. The Tianjic chip, on the other hand, is a digital approach to implementing neuromorphic computing and this makes it easier to use.

To prove that it can be trained and can do a complicated job - enter the riderless bike. Notice that this electric bike is driven, and kept upright by a Tianjic chip. It can accept voice commands and it has enough vision processing to follow a human:

To quote the introduction to the research paper:

"Using just one chip, we demonstrate the simultaneous processing of versatile algorithms and models in an unmanned bicycle system, realizing real-time object detection, tracking, voice control, obstacle avoidance and balance control."

The prototype of the chip was created in 2015 and development is ongoing. You can see that today's model includes an array of 156 "neurons" per chip:

Tianjic1

The FCore is like a digital simulation of a spiking neuron, but it can also be used to implement a non-spiking traditional neural network. The all-digital implementation gives it the flexibility to span both approaches:

Tianjic2

The development modules contain one chip or an array of 25 chips - the bike uses just one chip.

At this early stage it is difficult to say if this is a breakthrough or how important it is for the future. There are many other attempts to use neuromorphic chips - IBM's True North and Intel's Pohoiki Beach system to name just two. The importance of the Tianjic chip is that it is an all-digital system implementing hybrid networks and it indicates that China is making progress.
But it isn't necessarily going in the same direction.

Tianjic3

More Information

Towards artificial general intelligence with hybrid Tianjic chip architecture Paywalled 

 

Related Articles

Neuromorphic Supercomputer Up and Running

IBM's TrueNorth Rat Brain 

IBM's TrueNorth Simulates 530 Billion Neurons  

Flying Neural Net Avoids Obstacles

  

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on, Twitter, Facebook or Linkedin.

 

Banner


Python 2 End Of Life Threatens Security
28/08/2019

Python 2's end of life is fast approaching and the UK's National Cyber Security Centre has issued a warning to developers still clinging on to Python 2 of risks they will face as a result of no more b [ ... ]



Linux Kernel Turns 28 Today
25/08/2019

August 25th is taken to be the official birthday of the Linux. What's so special about 28? Well we managed to miss 21 and 25 so we are making sure we mark it this time around,


More News

graphics

 



 

Comments




or email your comment to: comments@i-programmer.info

 

 

Last Updated ( Sunday, 11 August 2019 )